The AI Street Journal: Safeguards, safer routes and recycled building parts
The former president wants a public framework for addressing AI’s economic effects and safety. Industry leaders are also discussing independent scrutiny of their models.
The briefing
Barack Obama urges Democrats to develop an AI safeguards plan. MIT researchers test a method for keeping generated answers within strict limits, while Atlas pursues factory partners for its recycled-plastic building system.
Obama urges Democrats to put AI safeguards on their agenda
Obama has called for Democrats to develop an AI safeguards plan if they regain the House. Separately, leading AI executives are discussing evaluator access and shared safety standards, rather than announcing binding rules.
At a Democratic fundraising event, House Minority Leader Hakeem Jeffries asked Barack Obama how congressional Democrats should approach AI. Obama called for a framework for public discussion if the party regains a House majority.
Anthony Ha’s TechCrunch report, drawing on New York Times coverage and a partial transcript supplied by Obama’s office, describes concerns about economic disruption and safety alongside potential benefits in drug development.
Jeffries endorsed the need for action. This is a call to develop policy, not a new law or a detailed set of obligations for companies.
Independent access enters the debate
Separately, Anthropic chief executive Dario Amodei proposed giving independent safety evaluators access to leading companies and models, alongside common safety standards between firms.
OpenAI chief executive Sam Altman responded positively, saying his company would commit to giving independent evaluators access comparable to employees. That commitment does not establish that a shared evaluation system is operating.
The political approaches also differ. The Trump administration’s legislative framework would override state laws and place responsibility for child safety on parents.
For companies building or using AI, the concrete questions are who gets access to test models and which level of government sets the rules. These proposals put both questions into the debate without settling either.
Market signal
MIT’s HardFlow tests a different route to enforcing AI constraints
MIT researchers report that HardFlow met every required constraint in their experiments while improving solution quality. The deployment-time method combines strict final requirements with goals such as shorter robot paths.
A robot route can be short and still hit an obstacle. MIT researchers have developed HardFlow to help generative models satisfy non-negotiable requirements while also improving qualities such as journey time.
The team comprises lead author Zeyang Li, Kaveh Alim and senior author Navid Azizan. Adam Zewe’s MIT News account describes the research, published in IEEE Transactions on Pattern Analysis and Machine Intelligence.
Common projection-based approaches force partial answers to meet constraints throughout generation. HardFlow instead steers that internal process while enforcing the hard requirements on the finished answer. The discarded intermediate samples need not themselves be usable solutions.
How the tests turned out
Using optimal-control methods, the algorithm divides a difficult optimisation problem into smaller steps. It can pursue an additional goal, such as reducing travel time, while keeping the final route clear of obstacles.
The team tested robot manipulation, navigation through mazes and image changes directed by text. They report that no required constraint was breached in those experiments and that the answers scored better than those from comparison methods.
HardFlow also generally needed no more computing time than its rivals, and sometimes less. The findings demonstrate results on those tasks, not a universal safety guarantee.
HardFlow offers developers a way to add strict output requirements to pretrained models without retraining them. In route planning, that could mean optimising travel time as well as avoiding obstacles, rather than choosing between the two.
What to watch
Atlas supplies recycled-plastic bridge parts as it pursues factory partners
Atlas supplied recycled composite trusses for a Massachusetts wetland bridge and is discussing international factory partnerships. Its waterless process turns shredded plastic and fibreglass into structural components using large-scale 3D printing.
A 40-foot bridge in a Massachusetts wetland uses Atlas’s recycled composite trusses, supplied to the US Army Corps of Engineers. Installation took less than a day. The company’s components also support structures including barns and docks.
Zach Winn’s MIT News report describes the business founded by MIT researcher A.J. Perez and former Maine senator Matt Pouliot. Their Atlas Factory Stack combines plastic recycling with robotic manufacturing and large-scale 3D printing.
The system melts shredded single-use plastic and combines it with American-made fibreglass, then prints building components. Atlas says the recycling step requires no water, reducing its dependence on local water supplies.
Factory partnerships remain in talks
The founders are discussing international deployments with prospective franchise partners. They want factories close to construction sites, using nearby plastic waste, rather than a single distant plant shipping components around the world.
Atlas puts a factory cell’s daily output at enough framing parts for roughly one small house. That measures structural components, not completed homes.
Perez also sees the possibility of producing parts more cheaply than injection moulding. It is a prospective saving, separate from the evidence of components already being used in construction.
For builders, Atlas offers another use for recycled plastic: structural parts rather than disposable goods. Its waterless process removes water consumption from that recycling step, while the bridge provides a concrete example of the components being used.
What to watch next
- For companies building or using AI, the concrete questions are who gets access to test models and which level of government sets the rules. These proposals put both questions into the debate without settling either.
- HardFlow offers developers a way to add strict output requirements to pretrained models without retraining them. In route planning, that could mean optimising travel time as well as avoiding obstacles, rather than choosing between the two.
- For builders, Atlas offers another use for recycled plastic: structural parts rather than disposable goods. Its waterless process removes water consumption from that recycling step, while the bridge provides a concrete example of the components being used.
The takeaway
For companies building or using AI, the concrete questions are who gets access to test models and which level of government sets the rules. These proposals put both questions into the debate without settling either.
The editor’s view
HardFlow offers developers a way to add strict output requirements to pretrained models without retraining them. In route planning, that could mean optimising travel time as well as avoiding obstacles, rather than choosing between the two.